UniLab is a free, open-source toolbox for students — think iLovePDF /
iLoveIMG, but with one radical difference: everything runs 100% inside your
browser. No uploads, no accounts, no ads, no task limits, no file-size caps,
and your documents never touch a server. Once loaded, it even works offline as
an installable PWA.
Built by students at Mae Fah Luang University for 1305493 Software
Engineering Case Studies (1/2569) — and for every student who has fought an
LMS upload cap at 11:55pm.
Why client-side?
| Upload-based sites | UniLab |
|---|
| Your transcript / ID scan | goes to their server | never leaves your device |
| Free tier | 2 tasks/day, size caps, ads, watermarks | everything, always |
| Slow campus Wi-Fi | upload + download round-trip | instant |
| Offline | ✗ | ✓ (PWA) |
| PDPA / privacy story | a policy you must trust | an architecture you can verify |
The trade-off, stated honestly: no server-grade conversions (PDF→Word needs a
server; we don't do it), and PDF compression rasterizes text. We think that's
the right trade for student daily life.
The tools (58)
| Category | Tools |
|---|
| 🖼️ Image (13) | Compress ("must be under X MB" target mode) · Resize · Crop (ID-photo presets) · Convert · Images→PDF · HEIC→JPG · Add Text · Rotate · Watermark · Remove Background · Blur Faces · Enlarge (Lanczos-3) · Photo Editor |
| 🎬 Video (7) | Compress · Trim · Convert · Resize for 9:16/1:1/16:9 · Video→GIF · Video→Photos · Screen Recorder |
| 🎵 Audio (11) | Enhance Voice (one-press clean-up) · Remove Noise (spectral + optional on-device neural) · Fix Volume (LUFS) · Change Speed (pitch-preserved) · Cut Silences · Equalizer (live preview) · Join Audio · Extract Audio · Trim · Convert · Voice Recorder |
| 📄 PDF (20) | Merge (page ranges + contents page) · Split (named parts) · Compress (3 levels or a size cap) · Rotate · PDF→Images · Organize · Watermark · Page Numbers · Sign · Crop · Edit · OCR (Thai + Burmese) · Fill Form · Unlock · Redact · Compare · PDF→Markdown · Scan to PDF · Word→PDF · Excel→PDF |
| ✍️ Text (2) | Word Counter (Unicode-correct — see below) · Citation Generator (APA 7 / MLA 9) |
| 🎓 Study (2) | GPA Calculator (Thai university scale, saved on-device) · Pomodoro Focus Timer |
| 🧰 Everyday (3) | Workflows · QR Code Maker (link + Wi-Fi, never expires) · Unit Converter (incl. Thai land units ไร่/งาน/ตร.วา) |
Workflows — chained tools, free
iLovePDF sells Workflows as a Premium feature, and it has to be paid for: every hop
in a chain costs them another upload, another store and another download. In a browser
the file is already in memory, so a chain is not only free — it is faster than
running the tools one at a time.
Pick your steps, save the chain, drop your files in. Ready-made ones cover the evenings
that actually happen: photos of notes → one small numbered PDF, lecture video → a
64 kbps MP3, report → stamped DRAFT and numbered.
The audio lab
Adobe's Podcast Enhance does its cleaning on Adobe's servers. UniLab's audio lab does
real signal processing in the tab: spectral noise gating that learns the room from the
gaps between sentences, mains-hum notching, LUFS loudness normalisation, pitch-preserving
speed change, silence cutting that tells you how many minutes it saves — and an optional
neural voice mode running RNNoise as WebAssembly, bundled with the site (112 KB,
same-origin, works offline). Honest limits, stated in the tools themselves: steady
background noise comes out; a voice drowned by a passing truck does not come back.
Video and audio, which no comparable site has
Neither iLovePDF nor iLoveIMG has a single video or audio tool. UniLab has eleven, built
on WebCodecs — the browser API that hands a page the same hardware decoder the video
player uses. That is how a 500 MB lecture recording gets trimmed in seconds without being
uploaded. We deliberately did not use ffmpeg.wasm: it would mean shipping ~31 MB of
WebAssembly and decoding on the CPU.
Your files are never stored — and you can watch the clock
Finished files live in this tab's memory with a countdown on screen, and are dropped when
it runs out, when you close the tab, or when you press the bin. Upload-based sites show
you the same countdown; the difference is that theirs is a promise about a copy on their
servers, and ours is about the only copy there has ever been.
Unicode done right 🇹🇭🇲🇲
Most counters get Southeast Asian scripts wrong. UniLab counts what you
actually see:
- Burmese ရွဲ့ = 1 character (not 4) — grapheme-cluster counting via
Intl.Segmenter, with a separate code points stat for portals that count
the raw way, and a combo inspector that shows the decomposition (ရ + ွ + ဲ + ့).
- Thai สวัสดี = 4 characters, and word counting works without spaces.
- Zero-width spaces (the invisible separators in Myanmar text) count as
spaces; ။ and ។ count as sentence ends.
- 👨👩👧👦 = 1 character, like your eyes say.
Run it locally
git clone https://github.com/mfu-hlaing/unilab.git
cd unilab
npm install
npm run dev
Vite + vanilla JavaScript — no framework to learn. Processing is done by
pdf-lib, pdf.js,
mediabunny (WebCodecs video/audio),
tesseract.js (OCR),
mammoth (.docx),
browser-image-compression,
cropperjs,
heic2any,
gifenc,
jszip and
qrcode — all in-browser.
Two tools download a model on first use — OCR fetches its language data, and Remove
Background fetches an ONNX model. Both are gated behind an explicit button that states
the size, and in both cases it is the engine that is downloaded: your file stays here.
What we deliberately don't do
PDF→Word, PDF→PowerPoint, PDF→Excel, HTML→PDF, password protection, PDF/A and repair
all need a server, so they are not here. Neither is an AI summariser or translator: those
send your document to a model provider, which is the exact thing this project exists to
avoid. Compress PDF rasterises pages, so its output has no selectable text — the tool
says so on screen.
Deploys to GitHub Pages automatically on every push to main
(workflow).
Contributing
New tools, translations (Thai/Burmese UI is on the roadmap), bug reports and
first-time contributions are all welcome — a new tool is ~1 file plus a
registry entry. Start with CONTRIBUTING.md; the one
unbendable rule is that everything must run client-side.
AI use disclosure
Prototype scaffolding and market research were AI-assisted (Claude), as
encouraged by the course's AI-first policy. All code is reviewed, tested in
the browser with real files, and owned by the team.
License
MIT — free to use, learn from, and fork.
